Xml grep

When working with xml, you'll often want to grep out an element that spans several lines. This can be hacked with awk or perl, but a more elegant solution is to use the parser in libxml2 (which is a requirement when installing apertium, so should be installed on your system already). This lets you use a simple version of XPath expressions to grep out full XML elements, without falling for the tempation to parse XML with regex.

Specifying the full path and the full pardef name:

$ xmllint --xpath '/dictionary/pardefs/pardef[@n="gen__apos"]' apertium-eng.eng.dix
<pardef n="gen__apos">
  <e>       <p><l/>          <r/></p></e>
  <e>       <p><l>'</l>         <r><j/>'<s n="gen"/></r></p></e>
</pardef>

But for dix files, it should be the same if you specify a relative path:

$ xmllint --xpath '//pardef[@n="gen__apos"]' apertium-eng.eng.dix
<pardef n="gen__apos">
  <e>       <p><l/>          <r/></p></e>
  <e>       <p><l>'</l>         <r><j/>'<s n="gen"/></r></p></e>
</pardef>

You can also search for substrings by using the 'contains' function:

$ xmllint --xpath '//pardef[contains(@n,"_adj")]' apertium-eng.eng.dix
<pardef n="expensive__adj">
  <e>       <p><l/>          <r><s n="adj"/></r></p></e>
</pardef>
<pardef n="ca__adj">…
# etc; gives all the adj pardefs

To get all c attributes:

$ xmllint --xpath '//@c' apertium-eng.eng.dix

To get c attributes only from <e> elements:

$ xmllint --xpath '//e/@c' apertium-eng.eng.dix

To get all attributes of the e element that has the lm "cake":

$ xmllint --xpath '//e[@lm="cake"]/@*' apertium-eng.eng.dix

To get the second dictionary section:

$ xmllint --xpath '/dictionary/section[2]/' apertium-eng.eng.dix

(or section[position()=2])

To count how many lm attributes (should equal how many lemmas) you have:

$ xmllint --xpath 'count(//e/@lm)' apertium-eng.eng.dix

Some corpora are formatted in XML and put e.g. the real text contents inside a particular element. Say the corpus puts all text inside <sentence> elements, you can grep them out with:

$ xmllint --xpath '*/sentence/text()' corpus.xml

But I want XML awk/sed/diff/patch/join/etc.!

To do more complex xml munging, you might want to install XML Starlet. The syntax takes a bit of getting used to, but it's quite powerful. If nothing else, it's worth getting because it can append newlines after each selection. E.g.

 $ xmllint --xpath '//e[@r="LR"]//text()' apertium-eng.eng.dix

would get text content of entries marked LR, but there will be no separator between each match. XML Starlet requires -a -few -more -options, but can output newlines after each match:

 $ xmlstarlet sel -t -m '//e[@r="LR"]//text()' -c . -n apertium-eng.eng.dix

"sel" means select/query, -t means output as text, -m means "match this XPath expression", "-c ." here means "output what was matched" and -n means add a newline.

Note that on some systems, the command for xmlstarlet is simply "xml".

I get "Unknown option --xpath"

You need to upgrade your libxml2 (--xpath was added in 2.7.7, back in Mar 15 2010).

External links

• http://oreilly.com/perl/excerpts/system-admin-with-perl/ten-minute-xpath-utorial.html